The reconstruction of plused magnetic flux leakage (PMFL) defects profiles is key for the quantitative study of defects characteristic. The used SVM kernel function can not approach to any object function in defects reconstruction. Aiming at this problem, based on the idea of multi-resolution approximation, the wavelet and the SVM are combine to form WSVM. Then，this method is introduced in 3-D defect reconstruction research in this paper. In the experiments, the horizontal component of magnetic flux density Bx is chosen as input data of WSVM nets, the defect geometric parameters: length, width and depth are output data. A mapping from PMFL response signals to 3-D profiles of defects was established, and the inversion of 3-D profiles of defects from magnetic flux leakage inspection signals was achieved. Experimental results show that the proposed method can combine the advantages of SVM and wavelet so that it obtains high precision and a good generalization ability, and capability of tolerating noise. It is a feasible method.